assign instance labels for all points in the scene :param points: (N, 3), points in world coordinate :param thres_connect: t

优化后的超点区域生长算法

Convert point-level positive integer data to histograms of segment-level labels, based on idx. :param item: 1D or 2D torch.Tensor :param

New pose using average position, z-axis, and up vector of input poses.

:param label0: (lr-1,), label distribuion of one seg :param labels1: (neigh,lr-1), label distribution of neighbors of the seg above

Convert torch.Tensor to numpy while respecting some constraints on output dtype. Integer tensors will be cast to the smallest possible integer

Cast an integer tensor to the smallest possible integer dtype preserving its precision.

Does part of what `Data.select()` does but in an ugly way. This is mostly intended for the `DataLoader` to be able to load `NAG` and s

Shallow copy of self. This may cause issues for certain types of downstream operations, but it saves time and memory. In practice, it

Makes sure all loggers closed properly (prevents logging failure during multirun).

Apply Turbo colormap to a normalized value tensor. For cuda acceleration, the colormap is preloaded as a tensor. Args: value (to

Helper function to compute the semantic, instance, panoptic segmentation metrics of a model on a given dataset, for given instance graph and p

Helper function to compute the semantic segmentation metrics of a model on a given dataset.

Creates videos out of the images saved to disk.

Compute the cross-product matrix of a vector k. Credit: https://github.com/torch-points3d/torch-points3d

Returns a string holding the current date and time. Useful for creating an output file or directory.

Runs a series of sanity checks on the attributes of self.

Sanity checks.

Sanity checks.

Convert a dense matrix to its CSR counterpart.

Compute consecutive unique identifiers for the edges. This may be needed for scatter operations.

Estimate the bounding sphere given camera pose

Estimate the centroid position of each object, based on the position of the clusters. Based on the hypothesis that clusters are relat

Estimate the centroid position of each target instance object, based on the position of the clusters. Based on the hypothesis that cl

points are XYZ RGB (RGB in 0-255), semantic label as nyu40 ids, instance label as 1-#instance, box as (cx,cy,cz,dx,dy,dz,semantic_label)

Perform TSDF fusion given a fixed depth range, used in the paper. voxel_size: the voxel size of the volume sdf_trunc

Experimental features, extracting meshes from unbounded scenes, not fully test across datasets. return o3d.mesh

Applies optional utilities before the task is started. Utilities: - Ignoring python warnings - Setting tags from command line - Rich

Search for the lowest point in each cell of a horizontal XY grid. This can be used to filter out points far from the ground, with some limita

Search for the points with low verticality. For verticality computation, see the `PointFeatures`. This can be used to filter out non-ground

Search for points within `threshold` Z-distance of the lowest point in the input cloud `xyz`. This can be used to filter out points far from

Overwrite torch_geometric from_data_list to be able to handle Cluster and InstanceData objects batching.

Generate an elliptical render path based on the given poses.

Generate an image with random ground truth and predicted instance and semantic segmentation data. To make the images realisitc, and to ensure

Parse a config file structure in search for .yaml files

Overwrite torch_geometric get_example to be able to handle Cluster and InstanceData objects batching.

mask: N, boolean mask level: int, choose certain level to get the related gaussian

:param points_label: (N, M), labels of all points in all views :param points_seen: (N, M), seen flag of all points in all vi

:params similar_meric: the metric to calculate the similarity between two primitives :params points_label: (N, M), labels of all poin

Reduce dimension of x based on PCA on the union of the n-simplex. This is a way of reducing the dimension of x while treating all input dimens

Index selection of pointers. Returns a new pointer tensor with updated pointers, along with an index tensor to be used to update any

Optionally recursive inlier search. The `xyz_query` and `xyz_search`. Search for points with less than `k_min` neighbors within a radius of `r

Method called when in-place reindexing groups. The group_indices are assumed to be sorted and group_indices[i] corresponds to the pos

The forward step will compute the partition on the instance graph, based on the node features, node logits, and edge affinities. The partition

Compute instance graph and per-edge affinity scores. :param edge_index: Tensor of size [2, num_edges] Edges connecting the cluste

Create a dictionary of torch_geometric composite transforms from a datamodule OmegaConf holding lists of transforms characterized by a `*trans

Create a transform from an OmegaConf dict such as: ```yaml transform: GridSampling3D params: size: 0.01 ```

Create a torch_geometric composite transform from an OmegaConf list such as: ```yaml - transform: GridSampling3D params:

对4x4变换矩阵进行插值,通过提取和插值四元数和平移向量。 插值是在每两个位姿之间进行的，并将所有的插值结果汇总为一个综合的结果。 :param matrices: 包含4x4变换矩阵的列表 :param times: 对应于每个位姿的时间戳列表

If self.has_edges, returns a boolean tensor of size self.num_nodes indicating which are absent from self.edge_index. Will raise an err